Lineage Lens
Track which lines of your code came from AI
Start new thread

The one provenance argument that actually creates urgency: you can't go back

by•

Something I have been thinking about in the AI governance space that I do not see discussed enough: provenance capture is not like most tooling categories.

With most observability or audit tooling, the reasoning is "we should have this so we're better positioned going forward." You can turn it on when the need becomes clear. You lose some history, but the tooling from that point forward is complete.

AI code provenance does not work this way.

The prompt a developer submits to Claude Code exists for a few hundred milliseconds in transit. After the model returns its response and the editor applies the change, that prompt is gone. Git records the diff. Nothing else records the origin by default. There is no reconstruct operation.

I built LineageLens partly because I ran into this problem in my own workflow — traced a bug to a block I had accepted from Cursor weeks earlier and had no way to recover what I had asked for. The architecture reflects this constraint directly: the Base extension (free, zero config, VS Code Marketplace) starts recording at the moment it installs. Not full provenance — file-level only until the proxy is running — but it starts preserving a timestamped

insertion history that cannot be recovered retroactively.

I am curious whether teams here have specifically run into the retroactive version of this problem: you had an incident or a compliance question, traced it to AI-generated code, and discovered the provenance window was already closed. How did that investigation go? Did it change how your team thought about when to start governance tooling?The VS Code extension is free and installs in under a minute if anyone wants to see what gets captured:

101 views

Add a comment

Replies

Best

That's a compelling point... most governance tools can be added later, but provenance is one of the few areas where missed history is lost forever. The cost of waiting may only become visible when you need answers and can't get them.

 Exactly. That irreversible timing aspect is what makes provenance fundamentally different from most observability tooling. With logs, metrics, or dashboards, delayed adoption usually means incomplete history. With AI provenance, delayed adoption can mean the original intent and generation context are unrecoverable forever.

The difficult part is that teams often do not feel the cost immediately. They feel it later during debugging, audit review, incident response, or compliance investigation — precisely when the missing context becomes impossible to reconstruct.

The idea that AI prompts disappear so quickly is something people do not fully realize until they actually need to trace a decision later.

 I think that realization surprises a lot of teams. People assume provenance can always be reconstructed later from Git history, logs, or editor state, but prompts and intermediate AI context are often extremely ephemeral. Once that moment passes, the evidentiary surface collapses very quickly.

That’s part of why I see provenance capture less as “historical analytics” and more as preserving a narrow operational window that disappears almost immediately after generation.

Interesting point about irreversibility. People assume everything in software can be reconstructed later from logs or commits, but prompts disappearing instantly creates a blind spot that traditional engineering practices were never designed for.

 That blind-spot framing is very close to how I see it too. Traditional engineering workflows evolved around artifacts that persist naturally: commits, diffs, logs, tickets, deployments. AI generation introduces a new layer where some of the most important context exists only briefly during inference and then disappears completely unless captured intentionally.

That creates a provenance gap that older software governance assumptions were never really designed to handle.

The you can't go back part is what stood out to me. A lot of tooling can be added later but missing history is missing history. Feels like one those things people only appreciate after getting burned once.

 I think that’s exactly the psychological shift. Most teams initially assume provenance is something they can “add later if needed,” because that’s how observability tooling usually works. The painful realization comes when they discover the original AI context was transient and never persisted anywhere.

By the time the need becomes obvious, the evidentiary window may already be closed.

Drop the comments below !!